Convex conjugate¶

The supremum transform mapping an extended-real function on a vector space to the greatest affine lower-bound gap over its dual space.

Version: v2 · 2026-09-19 · History
Domain-specific #: 3903
Domain group: Formal Sciences
Origin domain: Mathematics
Subdomain: Convex Analysis → Mathematics
Aliases: Legendre–Fenchel transform, Fenchel conjugate

Core Idea¶

For f, its Fenchel conjugate f-star at a dual vector is the supremum of the dual pairing minus f; the result is convex and lower semicontinuous even when f is not. Every primal point supplies an affine function on the dual, their pointwise supremum forms the conjugate and repeating the transform yields the closed convex envelope under Fenchel-Moreau hypotheses. The abstraction is therefore identified by a declared carrier, a transformation or constraint over that carrier, and an invariant that tells an analyst whether the named structure is genuinely present.

How would you explain it like I'm…

Sliding Stick Picture

Draw a curvy line, and hold a straight stick tilted at some angle below it. Slide the stick up until it just touches the curve. If you write down, for every tilt, how far up the stick had to go, you get a new description of the curve. This description can't see into any dents in the curve, so if you turn it back into a curve, the dents come out filled in.

Best Line for Every Slope

Take a function, drawn as a curve. For every possible slope, imagine the highest straight line with that slope that still stays below the curve everywhere. Writing down where each such line sits, slope by slope, gives a new function called the convex conjugate. The new function is always bowl-shaped (convex), even if the original curve had bumps and dips. If you do the process twice, you get back the curve with its dents filled in, like the curve you'd get by stretching a sheet up underneath it.

Fenchel Conjugate Transform

The convex conjugate (or Fenchel conjugate) turns a function f into a new function f*. For each slope y, f*(y) is the largest value of y·x minus f(x) over all x: how far the straight line y·x can rise above the graph of f. Each point x of the original gives a straight line in y, namely y·x - f(x), and f* is the upper envelope (the pointwise maximum) of all those lines. Because a maximum of straight lines is always convex and has no downward jumps, f* is convex and lower semicontinuous even when f is not. Applying the conjugate twice gives f**, which under the right conditions (the Fenchel-Moreau theorem) is the closed convex envelope of f: the largest convex, lower semicontinuous function lying below f. If f was already convex and lower semicontinuous, you get f back exactly.

For an extended-real-valued function f on a real vector space X paired with its dual X*, the Fenchel conjugate is f*(y) = sup over x of (<y, x> - f(x)). Each primal point x contributes an affine function y -> <y, x> - f(x) on the dual, and f* is the pointwise supremum of this family, so f* is convex and lower semicontinuous whether or not f is. The biconjugate f** = (f*)* is the closed convex envelope of f, the largest lower semicontinuous convex minorant, under the hypotheses of the Fenchel-Moreau theorem; in particular f** = f exactly when f is proper, convex, and lower semicontinuous (or identically plus or minus infinity). The definition requires fixing the topology, the continuous dual and pairing, properness conventions, the domain of the supremum, and the conjugate convention used, since variants (for example sign or concave conventions) differ.

Scope of Application¶

Convex conjugate belongs to convex analysis and is useful where the analyst can specify the typed convex analysis carrier, including objects, relations, parameters, conventions, evidence, boundaries, and comparison targets, then evaluate the real vector space and topology, continuous dual and pairing, extended-real function and properness, supremum domain, conjugate convention, lower-semicontinuity and convexity, biconjugate hypotheses and equality cases are explicit. The scope is broad within that domain but bounded by the need for the real vector space and topology, continuous dual and pairing, extended-real function and properness, supremum domain, conjugate convention, lower-semicontinuity and convexity, biconjugate hypotheses and equality cases are explicit. The entry records a descriptive analytical identity; practical use requires the governing domain's evidence, standards, and safety obligations.

Clarity¶

The abstraction clarifies a crowded vocabulary by making the real vector space and topology, continuous dual and pairing, extended-real function and properness, supremum domain, conjugate convention, lower-semicontinuity and convexity, biconjugate hypotheses and equality cases are explicit the center of the account. A claim should name the carrier, the governing operation or relation, the applicable assumptions, and the recognition test.

Manages Complexity¶

Without the abstraction, an analyst must reason directly over many local details: the carrier roles, admissibility assumptions, competing conventions, derived invariants, boundary cases, and proof or validation obligations specific to Convex conjugate. Convex conjugate compresses them into the roles in the structural signature. That compression permits comparison across instances without erasing the variables that determine validity. It also exposes which details may be varied safely and which are constitutive.

Abstract Reasoning¶

Identify the carrier. State what the elements, states, objects, or observations are: the typed convex analysis carrier, including objects, relations, parameters, conventions, evidence, boundaries, and comparison targets. Reject examples whose alleged carrier belongs to a different problem.
Lock the constitutive rule. Express the real vector space and topology, continuous dual and pairing, extended-real function and properness, supremum domain, conjugate convention, lower-semicontinuity and convexity, biconjugate hypotheses and equality cases are explicit independently of one notation or implementation.

Knowledge Transfer¶

Knowledge transfers strongly among subfields of convex analysis because they reuse the typed convex analysis carrier, including objects, relations, parameters, conventions, evidence, boundaries, and comparison targets, Every primal point supplies an affine function on the dual, their pointwise supremum forms the conjugate and repeating the transform yields the closed convex envelope under Fenchel-Moreau hypotheses., and type the carrier, state every parameter and convention in the definition, test that the real vector space and topology, continuous dual and pairing, extended-real function and properness, supremum domain, conjugate convention, lower-semicontinuity and convexity, biconjugate hypotheses and equality cases are explicit, compare the nearest accepted identity, and report counterexamples, uncertainty, and limiting cases.

Relationships to Other Abstractions¶

Current abstraction Convex conjugate Domain-specific

Parents (1) — outgoing typed relations

Convex conjugate is a kind of Duality Prime

It remains its own entry because its identity is fixed by the real vector space and topology, continuous dual and pairing, extended-real function and properness, supremum domain, conjugate convention, lower-semicontinuity and convexity, biconjugate hypotheses and equality cases.

Children (1) — abstractions linked here by typed relations

Cramér's Theorem (Large Deviations) Domain-specific presupposes Convex conjugate

Cramér's rate is defined as the convex conjugate of the one-observation log MGF.

Hierarchy path (1) — routes to 1 parentless root

Convex conjugate → is a kind of Duality

Neighborhood in Abstraction Space¶

Convex conjugate sits in a crowded region of the domain-specific corpus (8^th percentile for distinctiveness): several abstractions share nearly its structure, so a description that fits it tends to fit its neighbors too.

Family — Convex Analysis & Optimization Structures (8 abstractions)

Nearest neighbors

Indicator function (convex analysis) — 0.95
Convex hull — 0.93
Supporting hyperplane — 0.92
Differentiable vector-valued functions from Euclidean space — 0.92
Linear matrix inequality — 0.92

Computed from structural-signature embeddings · 2026-10-25